🛠️ UberChatMaster: Take Control with Your Own AI Interface

[Uncle Gizmo]

Why an unreliable AI tool is pushing me to build something better — and how it may become much more than I planned
By Tony @Uncle Gizmo

Over the past year, I’ve been experimenting with how large language models (LLMs) can assist developers — not just with snippets of code, but with building real software.

My background is in Microsoft Access and VBA, but recently, I found myself exploring Python, Flask, JavaScript, and web-based architecture. The catch? I didn’t sit down to learn Python from scratch. I used AI tools — like ChatGPT, Grok, and Gemini — as collaborators. And the result was something I call UberChatMaster: a multi-model AI chat platform that runs locally and gives me total control over my interactions.

But a recent experience reminded me just how important control really is.

Gemini Advanced 2.5 Pro: Impressive and Capable​

Let me start with the good: Gemini Advanced 2.5 Pro is a highly capable language model. In my experience, it’s solved complex coding and logic issues that even ChatGPT and Grok couldn't. It’s sharp, fast, and contextually aware.

If you’re looking for a strong AI coding partner, Gemini Advanced deserves serious attention.

Google AI Studio: The Weak Link in the Chain​

The trouble came not from the model, but from the Google AI Studio interface — the web platform at aistudio.google.com meant to interact with Gemini.

At the time I used it (May 2025), I ran into several frustrating issues:

• Chat titles were changed without my input
• Entire conversations disappeared with no way to retrieve them
• Context loss and disjointed continuity over longer sessions
• When I asked Gemini directly, it confirmed: “I cannot rename chats or delete conversations” — meaning these were platform-level problems

To be fair, Google clearly states that AI Studio is meant for experimentation, not production use. But even with that in mind, it simply wasn’t reliable enough for me to work effectively. Losing access to past discussions — is unacceptable.

---

Why I Built UberChatMaster​

That experience highlighted the value of something I’d already been building: UberChatMaster.

UberChatMaster from Nifty ChatGPT​

Originally just an experiment to unify multiple LLMs behind one interface, it’s now turned into something much more useful — and potentially much more powerful.

• Every message — prompt and response — is stored in a local SQLite database
• I can switch between OpenAI, Ollama, and soon Grok and Gemini (via API)
• I control the data, the model selection, and the conversation history
• Nothing gets lost unless I choose to delete it

It’s not fancy, but it’s rock solid. And it works the way I need it to.

---

A Surprising Realization​

Here’s where things get interesting.

Over the years, I’ve been saving my side of AI conversations using speech-to-text logs — a habit I picked up long before UberChatMaster existed. Now I realize: those are my notes. That’s my thinking — my decision trees, design patterns, prompts, and bug fixes — all sitting in raw form.

Now, with UberChatMaster capturing both sides of my AI conversations, I’m effectively logging my intellectual process in a searchable, structured way. With a bit more code — and maybe a vector database — I could start searching, analysing, and building upon my own historical interactions.

That’s when I thought of Ray Kurzweil
Kurzweil, the inventor and futurist, has long spoken about building a digital version of himself — an AI trained on his life’s emails, notes, and conversations. Not an abstract AI assistant, but one trained on his actual mind.

I’m not saying UberChatMaster is anywhere near that. But I am saying: it could be the seed of something like it. And I suspect I’m not the only developer who could benefit from that kind of personal memory assistant — built on their own thought history, not scraped data from the internet.

---

Want to See How I Did It?​

I’m creating a short video series to show other experienced developers — especially those from an Access or VBA background — how I built UberChatMaster by collaborating with AI, not learning everything from scratch.

[Pre-order the series on Gumroad for $10 here.]
If 10 people back the project, I’ll release the full series and expand it into something much bigger.

You’re not buying UberChatMaster — you’re learning how I built it using LLMs, prompting, debugging, and architectural thinking.

---

Final Thoughts​

Gemini Advanced 2.5 Pro is impressive. But Google AI Studio, as of today, just isn’t ready for serious dev work.

Meanwhile, the tool I built for my own use — UberChatMaster — is already solving problems I didn’t expect.
Not just switching LLMs, not just saving prompts…
…but maybe even helping me build a thinking partner shaped by my own past work.

It’s early days. But it’s worth exploring. ---